Legion warms to task of solving icy mystery
6 February 2017
Legion, a key component in UCL's world-class suite of computing platforms supported by Research IT Services (RITS), has helped a team of researchers from UCL and Germany to better understand how ice crystals form in the clouds.
This well-established, constantly evolving cluster provided the power, speed, ready availability and easy use vital to unlocking secrets surrounding a process that underpins 90% of rain formation.
By performing many billions of complex calculations in just a matter of weeks, Legion enabled the team to achieve a remarkable breakthrough in rapid time and finally reveal new insights about ice production in the atmosphere.
Rain: A Riddle
In terms of affecting the lives of everyone on the planet, few issues are as far-reaching as the climate. But much remains unknown or unclear. Fresh insights are needed to help us respond to the climate more effectively and predict it more accurately. Take rain. The basic process giving rise to rain is well understood: ice crystals form on the surface of dust-specks high in the atmosphere. But why does ice only form on a tiny fraction of such particles?
To solve the puzzle, working with Karlsruhe Institute of Technology (KIT) and funded by the European Research Council and the Royal Society, UCL's Interfaces Catalytic and Environmental (ICE) Group focused on how ice forms on feldspar - a mineral common in the Earth's crust that air currents whip up into the atmosphere when eroded.
"We needed to discover what was happening at a molecular level," says PhD student Philipp Pedevilla, who worked on the project with Professor Angelos Michaelides. "High-resolution electron microscope images of actual particles only take you so far. We had to explore the effect on ice formation of thousands of possible ice structures on feldspar particles."
The solution was to run computer simulations. But while the team had the expertise to develop suitable software code, running the required number of simulations on standard desktops would take years. RITS, though, had a facility perfectly equipped to meeting the demands of this classic number-crunching challenge.
Legion: Tried and Tested
" Legion delivers both the compute power and the sheer adaptability essential to researchers, whatever their field." - Dr Owain Kenway, RITS
Hosted at UCL's Bloomsbury data centre, Legion has evolved relentlessly to keep pace with the increasing size and complexity of demands placed on it. "The cluster currently comprises over 7500 Intel x86-64 CPU (Central Processing Unit) cores spread over a wide variety of node types in order to solve different kinds of problem," says Dr Owain Kenway, who leads the Research Computing Applications and Support Team at RITS. "So it delivers both the compute power and the sheer adaptability essential to researchers, whatever their field."
Specifically, Legion provides three key benefits to UCL researchers:
- Fast - It copes smoothly with computing challenges, including CPU-based parallel processing where jobs are broken down into small calculations that run simultaneously and produce results much quicker than if they ran in series. Legion, then, can complete big, complex tasks within practical timeframes.
- Flexible - equipped with a heterogeneous architecture that ensures outstanding versatility, Legion uses the same Operating System and applications as other UCL platforms - making it easy to switch between them - and incorporates a common software stack so users don't need to make big changes to the codes they run.
- Free - there's no direct charge at the point of use, access is simple and straightforward, and utilisation is supported by free help, advice and training from RITS, as requested and required. Using Legion avoids the need to apply for national supercomputing resources with limited availability and long waiting times.
"Any permanent UCL staff member can apply for an account," Dr Kenway explains. "Once they've got one, they can log in remotely and, using the simple guidance we provide, parcel up and submit jobs for us to schedule."
Right as Rain
" Legion dealt comfortably with the detailed molecular-scale simulations we needed to run." - Philipp Pedevilla, ICE Group, UCL
The investigation harnessed Legion over a period of 6 weeks. In total, it used over 8095 hours of CPU time - effectively meaning that the task would have taken a desktop, running flat-out continuously, almost a year to complete. "High performance computing is a key component of our work in general, so the advice we needed during this project really focused on getting optimum effectiveness from the specific code we'd written for it," Philipp Pedevilla says. "The streamlining of the whole process and the way Legion readily and rapidly copes with big workloads is always impressive. It dealt comfortably with the detailed molecular-scale simulations we needed to run".
The results were fascinating. They revealed that, where the surfaces of feldspar particles were smooth, ice crystals couldn't and didn't form. In fact, they only did so where microscopic splits, fissures or other defects randomly occurred on the material's surface - finally explaining why ice doesn't routinely form on every feldspar particle.
"This discovery is a key piece in the jigsaw and will ultimately help improve the ability to predict rain formation in a changing climate," comments Philipp. "By helping us assess many thousands of possibilities so quickly, Legion enabled us to publish our findings much faster than we could otherwise have achieved. The facility will also be key as we extend our investigations to multiple other classes of mineral. That's a huge task, of course, but the availability of Legion and UCL's other platforms means we've got exactly the right tools for tackling it."
- Snowflake (Credit: Michael Peres)
- A combination of quantum mechanical calculations and scanning electron microscopy experiments reveals how ice crystals can form in clouds (Credit: Philipp Pedevilla)